Surfaces∗

(COMS 4770/5770 Notes)

Yan-Bin Jia

Nov 21, 2024

1 Definition
To define a surface, we need the concepts of continuity and homeomorphism of mapping from Rm
to Rn . Let X and Y be subsets of Rm and Rn , respectively. A map f : X → Y is continuous
at a point p ∈ X if points in the set near p are mapped by f onto points in Y near f (p). More
precisely, f is continuous at p if, for any ǫ > 0, there exists δ > 0 such that kf (u) − f (p)k < ǫ for
any u ∈ X with ku − pk < δ. The function is continuous if it is continuous at every point of X.
A map f : X → Y is one-to-one (or injective) if f (p) 6= f (q) whenever p 6= q. The map is onto
(or surjective) if for every y ∈ Y there exists some x ∈ X such that f (x) = y. A map is called
bijective if it is both one-to-one and onto. If f is bijective, there exists a function from Y to X
called the inverse of f and denoted by f −1 . It is defined as f −1 (y) = x whenever f (x) = y.
If f : X → Y is continuous and bijective, and if its inverse map f −1 : Y → X is also continuous,
then f is a homeomorphism and the two sets X and Y are homeomorphic.
A subset S of R3 is a surface if, for every point p ∈ S, there is an open set U ⊆ R2 and an open
set W ⊆ R3 containing p such that S ∩ W is homeomorphic to U .

p S

A surface is thus viewed as a collection of homeomorphisms σ : U → S ∩ W , which are called

surface patches. The collection is called the atlas of the surface. Every point of the surface lies on
at least one surface patch in the atlas.
∗
Most of the material is adapted from [3].

1
Example 1. Plane Every plane in R3 is a single surface patch. Let p be a point in the plane, and a and b
two orthogonal unit vectors parallel to the plane. Then any point q in the plane can be represented as

q = p + ua + vb.

Hence, the surface patch is σ(u, v) = p + ua + vb with the inverse map σ −1 (q) = ((q − p) · a, (q − p) · b).

Example 2. Monge patch Let f be a differentiable real-valued function on an open set D ⊆ R2 .

The graph of f , i.e., the set of all points in R3 whose coordinates satisfy the equation z = f (x, y) is a
surface referred to as the Monge patch. The figure below plots the paraboloid z = x2 + y 2 over the domain
{(x, y) | x2 + y 2 ≤ 1}.

Example 3. Sphere The unit sphere

S 2 = {(x, y, z) ∈ R3 | x2 + y 2 + z 2 = 1}

is a surface.

z z z
π
σ̃
σ
σ(θ, φ)
C˜
θ C
y y y
φ π
2

x x x

We first consider the most common parametrization using latitude θ ∈ [− π2 , π2 ] and longitude φ ∈ [0, 2π]:

σ(θ, φ) = (cos θ cos φ, cos θ sin φ, sin θ). (1)

This is shown in the left figure above. However, there are two issues. First, σ is not injective over [− π2 , π2 ] ×
[0, 2π]. Second, this Cartesian product is not an open subset of R2 and hence cannot be used as the domain
of a surface patch. We consider the largest open set:
n π π o
U = (θ, φ) − < θ < , 0 < φ < 2π . (2)
2 2

2
Now, the image σ(U ) is the sphere minus the great semi-circle C which, shown in the middle figure above,
consists of the points of the form (x, 0, z) with x ≥ 0. Thus σ defines only a patch of the sphere.
To show that the sphere is a surface, we need to construct at least one more surface patch to cover the
semi-circle C omitted by σ. Imagine rotating σ by π about the z-axis and then by π/2 about the x-axis.
The new configuration of the great semi-circle, shown as C˜ in the right figure above, will not overlap with
its original configuration C. Essentially, we define a new patch σ̃ : U → R3 given by
 
π cos θ cos φ
σ̃(θ, φ)⊤ = Rotx Rotz (π)  cos θ sin φ 
2
sin θ
   
1 0 0 −1 0 0 cos θ cos φ
=  0 0 −1   0 −1 0   cos θ sin φ 
0 1 0 0 0 1 sin θ
 
− cos θ cos φ
=  − sin θ . (3)
− cos θ sin φ

It is the clear that the two great semi-circles C and C˜ not covered by σ and σ̃, respectively, do not interest.
The two patches thus form an atlas of the sphere.

Often in an application, multi-

z z ple patches need to be used. For
instance, consider that the sphere
σ σ̃ from Example 3 is rolling on a hori-
zontal plane. Its contact point with
C˜ the plane will trace out a (green)

p y y curve on the sphere as shown in part

p (a) of the left figure, where the x-
C y-z frame is rigidly attached to the
sphere. The first patch parametriza-
x x tion (1) is used to describe the
motion σ(θ(t), φ(t)) of the contact
(a) (b)
point in terms of time t on the
sphere. When the contact reaches a position p near the half great circle C, the value φ is close to 0.
Namely, (θ, φ) is near the boundary of the domain U given in (2). At this point, we want to switch
to the second patch σ̃ defined in (3) by solving for the new values of θ and φ from the following
equation:  
− cos θ cos φ
p= − sin θ .
− cos θ sin φ
As shown in (b), the point p is very far from the half great circle C˜ that is not described by the
patch σ̃.
A unit vector û ∈ R3 can be viewed as a point on the unit sphere, and therefore parameterized
with φ and ψ to lie in one of the above two patches σ(φ, ψ) and σ̃(φ, ψ). This q eliminates the
redundancy of representing û as a vector (ux , uy , uz )⊤ , on which the following u2x + u2y + u2z = 1
needs to be imposed.

3
 Recall from an earlier lecture that any rotation can be represented by a unit quaternion:
α α
q = cos + û sin ,
2 2
where the unit vector û gives the direction of the axis of rotation, and α ∈ [0, π) is the angle of
rotation. We can thus parameterize q using three parameters (α, φ, ψ), where φ and ψ parametrize û
on the unit sphere. Clearly, such parametrization has no redundancy. Suppose q(t) represents
the changing orientation of a rigid body with time under the influence of some external force
(e.g., the gravitational force). The orientation can be conveniently tracked by updating the three
functions α(t), φ(t), and ψ(t) — often through integrations of some equations derived using rigid
body dynamics.

2 Smooth Surface
A map f = (f1 , f2 , . . . , fn ) from an open subset of Rm to Rn is smooth if each component fi ,
1 ≤ i ≤ n, has continuous partial derivatives of all orders. A surface patch σ : U → R3 is regular if
it is smooth and the vectors σu = ∂σ/∂u and σv = ∂σ/∂v are linearly independent at every point
(u, v) ∈ U . Equivalently, the vector product σu × σv 6= 0 at every point of U . A smooth surface is
a surface σ whose atlas consists of regular surface patches. It is clear that the plane in Example 1
is a smooth surface.

Example 4. For the unit sphere in Example 3, it is obvious that σ and σ̃ are smooth functions. To verify
regularity, we obtain

σθ = (− sin θ cos φ, − sin θ sin φ, cos θ), (4)

σφ = (− cos θ sin φ, cos θ cos φ, 0). (5)

Thus
σθ × σφ = (− cos2 θ cos φ, − cos2 θ sin φ, − sin θ cos θ),
which has norm kσθ × σφ k = | cos θ|. Since on this patch, − π2 < θ < π
2, so cos θ 6= 0. Similarly, we can verify
that σ̃ is also regular.

3 Tangent Plane and Orientability

If a (smooth) curve γ : (a, b) → R3 is contained in a surface patch σ(U ) in the atlas of a surface S,
there exists a map t 7→ (u(t), v(t)) such that

γ(t) = σ(u(t), v(t)). (6)

Here the functions u and v are smooth. Conversely, if u and v are smooth, then (6) defines a curve
lying in S.
The tangent space at a point p of a surface S consists of the tangent vectors of all curves in S
that pass through p. Suppose p = σ(u0 , v0 ).

Theorem 1 The tangent space is spanned by the vectors σu (u0 , v0 ) and σv (u0 , v0 ).

4
Proof Let γ(t) = σ(u(t), v(t)) be a smooth curve in S. Denoting differentiation with respect
to t by dot, we have
γ̇ = σu u̇ + σv v̇.
Namely, γ̇ is a linear combination of σu and σv .
Next, we show that every vector spanned by σu and σv is the tangent vector at p of some curve
in S. Such a vector is of the form ξσu + ησv , for some ξ and η. Consider the smooth curve

γ(t) = σ(u0 + ξt, v0 + ηt).

At t = 0, i.e., at the point p, we have

γ̇ = ξσu + ησv .

Since σu and σv are linearly independent, the tangent space is indeed the tangent plane, which
by definition is independent of the patch choice. The tangent plane is uniquely determined by a
unit normal to S at p, which is perpendicular to the tangent plane. Though there are two such
vectors, choosing a surface patch σ leads to a definite choice, that is,
σu × σv
nσ = .
kσu × σv k

The above normal is called the standard unit normal.

Unlike the tangent plane, the normal nσ is not quite independent of the choice of the patch σ.
An orientable surface S has a canonical choice of unit normal at every point, obtained by taking
the standard unit normal of each surface patch in the atlas of S. Most of the surfaces we shall
discuss are orientable. Example 5 describes one that is not.

Example 5. The Möbius band is obtained by cutting a closed band into a single strip, giving one of the
two ends thus produced a half twist, and then reattaching the two ends.

We omit the details in showing that the Möbius band is not orientable according to the definition. An
intuitive explanation is given by considering a closed path in the surface that starts at point p, as shown in
the above figure. Whichever of the two unit normal vectors at p is chosen at the start, the normal vector
will be in the opposite direction when returning to p at the end of the loop.

5
4 Surface in Implicit Form
A surface may also be given in an implicit form f (x, y, z) =
0. As an example, Euler’s quartic surface x4 + y 4 + z 4 = 1
is plotted in the figure on the right.1 The implicit function
theorem [1] implies that the surface f (x, y, z) = 0 can be
parametrized locally at a point p which has non-vanishing
gradient, i.e.,
 
∂f ∂f ∂f
∇f |p = , , 6= 0.
∂x ∂y ∂z
p

Let σ(u, v) be a local parametrization at p. A smooth

curve γ(t) = (x(t), y(t), z(t)) in the surface passing by p
satisfies (6) for some functions u(t) and v(t). Differentia-
tion of f (x, y, z) = 0 with respect to t yields

∂f dx ∂f dy ∂f dz
0 = · + · + ·
∂x dt ∂y dt ∂z dt
= ∇f · (ẋ, ẏ, ż)
= ∇f · γ̇
= ∇f · (σu u̇ + σv v̇).

Since the curve γ(t) is arbitrary, the gradient ∇f is orthogonal to σu and σv at p, following the
reasoning used in the proof of Theorem 1. In other words, ∇f is orthogonal to the tangent plane;
hence it is a surface normal at p.

5 More Surface Examples

Example 6. Generalized cylinder This surface is obtained by
translating a curve γ : (c, d) → R3 in the direction given by the unit
vector a. Translating the point γ(u) by the vector va parallel to a
yields a new point
σ(u, v) = γ(u) + va,
N
which is the parametrization of the surface. Thus σ : (c, d) × R → R3 . Π
The two partial derivatives are σu = γ̇ and σv = a.

Example 7. Ruled surface A ruled surface is a surface formed by straight lines, called the rulings of the
surface. Let α(u) be a curve in R3 that meets every such line. Thus every point p in the surface must lie
on one of the straight lines that intersects α at, say, q. Suppose q = α(u) and let δ(u) be a vector in the
direction of the line through α(u). Then p has the form

σ(u, v) = α(u) + vδ(u).

for some v ∈ R.
1
This is an algebraic surface because f is a polynomial.

6
 p ruling
q
δ

The hyperbolic paraboloid is defined implicitly as

x2 y2
z= 2
− 2,
a b
where a, b > 0. It can also be parametrized as
γ(u, v) = (a(u + v), bv, u2 + 2uv).
The next figure shows the rulings on the hyperbolic paraboloid with a = 3 and b = 3 when viewed from the
point (0, 9, 0) on the y-axis.

Example 8. Surface of Revolution A surface of revolution is generated by rotating a plane curve

around a straight line in the same plane. The curve is referred to as the profile curve and the line as the axis
of rotation. Every circle generated by rotating a fixed point on the profile curve around the rotation axis is
called a parallel of the surface. Every curve on the surface obtained by rotating the profile curve through a
fixed angle is called a meridian.
Let the axis of rotation be the z-axis and the plane containing
the profile curve to be the xz-plane. Any point p of the surface is
z obtained by rotating some point q on the profile curve through an
angle v around the z-axis. Let the profile curve be parametrized as
meridian
γ(u) = (f (u), 0, g(u)).
parallel Then p is given as
v
σ(u, v) = (f (u) cos v, f (u) sin v, g(u)).
γ(u)
Now we compute the two tangent vectors as partial derivatives:
y
σu = (f˙ cos v, f˙ sin v, ġ),
σv = (−f sin v, f cos v, 0).
x
7
We have

σu × σv = (f ġ cos v, −f ġ sin v, f f˙),

q
kσu × σv k = f 2 (f˙2 + ġ 2 ).

Thus σu and σv are linearly independent if f (u) 6= 0, that is, if γ(u) does not lie on the z-axis, and if f˙(u)
and ġ(u) are not simultaneously zero. If the two vectors are linearly independent at every point, then we
might as well assume f (u) > 0 and view the function as the distance of σ(u, v) from the axis of rotation.

References
[1] J. R. Munkres. Analysis on Manifolds. Reading, MA: Addison-Wesley, 1991.

[2] B. O’Neill. Elementary Differential Geometry. Academic Press, Inc., 1966.

[3] A. Pressley. Elementary Differential Geometry. Springer-Verlag London, 2001.